Curvature scale space (CSS) descriptor is a robust multiscale approach for closed shape representation, which was one of the features selected to describe objects in the MPEG-7 standard. In this work, a novel method to symbolic representation for curvature scale space descriptor is presented to improve its performance. The canonical CSS descriptor is firstly mapped to circular vector map, in which, the angle and the Gaussian standard deviation are normalized and quantified to equal interval, respectively. Thus the CSS descriptor of every image can be converted to a symbol string used as shape feature. And then the distance index table is established, by which, the distance between any two CSS descriptor in circular map can be obtained directly rather than by using complicated formula. Comparative experimental results show that our proposal performs faster, simpler and more effective than canonical CSS matching. Furthermore, most traditional retrieval algorithms based on strings may be used to shape retrieval as well.